model-writtenrun 9133 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jeff Davis proposed password_protocol to let libpq reject password authentication weaker than a client-selected threshold, initially ordering plaintext, MD5, SCRAM, and SCRAM with channel binding #1.
- Michael Paquier questioned the hierarchy's extensibility and requested authentication and SSL tests #2, while Stephen Frost emphasized the separate needs to avoid revealing passwords and to require channel binding against interception #5.
- Heikki Linnakangas proposed rules combining authentication restrictions with transport security #8, but Jeff cautioned that complex client lists would be awkward in connection strings and could break across upgrades because most authentication methods cannot be negotiated #10.
- Peter Eisentraut challenged how a hierarchy would rank future SCRAM hashes and favored an explicit list covering all authentication methods #15 #20, while Jonathan S. Katz initially supported a documented minimum-strength hierarchy #14.
- Tom Lane asked which concrete security problem required new client controls #22; Stephen explained the channel-binding downgrade attack #23, and Peter separated that problem from password disclosure and unencrypted transport #24.
- Jeff narrowed the proposal to channel_binding=prefer|require, arguing that it addressed the immediate security need without committing to a general authentication-policy language #29; Stephen supported proceeding while leaving broader client controls for future work #33, and Michael accepted that direction #35.
- Jeff's replacement patch implemented the channel_binding parameter #36, and Michael's review requested disable mode, a matching default for builds without SSL, early rejection of incompatible authentication requests, and positive and negative tests #37.
- Jeff incorporated those requests, including SCRAM messages that correctly advertise disabled channel binding #38; Michael then found that setting channel_bound at SASL initialization let a rogue server send AUTH_REQ_OK before completing the exchange #39.
- Jeff added SASL-completion tracking, improved diagnostics, documented PGCHANNELBINDING, and added an MD5 rejection test #40; Michael proposed checking the existing SCRAM state instead of maintaining duplicate flags #41, which Jeff implemented as pg_fe_scram_channel_bound() #42.
- Michael supplied a fix-up that clarified state checks, relocated non-SSL tests to the password suite, added invalid-option coverage, and tidied formatting, marking the patch ready for a committer #43; Jeff folded it into the final submission and added a certificate-authentication rejection test and clearer comments and diagnostics #44.
- Michael reviewed the final submission and requested only capitalization of SCRAM in a comment #45; Jeff committed that version with the correction as d6e612f837.
- The thread ended with Michael confirming the commit and noting harmless indentation noise #46; channel-binding controls had landed, while the broader password-protocol and authentication-list proposals remained outside the chosen scope.